The Lenski et al 2003 paper, The evolutionary origin of complex features, is really worth reading. Here’s the abstract:
A long-standing challenge to evolutionary theory has been whether it can explain the origin of complex organismal features. We examined this issue using digital organisms—computer programs that self-replicate, mutate, compete and evolve. Populations of digital organisms often evolved the ability to perform complex logic functions requiring the coordinated execution of many genomic instructions. Complex functions evolved by building on simpler functions that had evolved earlier, provided that these were also selectively favoured. However, no particular intermediate stage was essential for evolving complex functions. The first genotypes able to perform complex functions differed from their non-performing parents by only one or two mutations, but differed from the ancestor by many mutations that were also crucial to the new functions. In some cases, mutations that were deleterious when they appeared served as stepping-stones in the evolution of complex features. These findings show how complex functions can originate by random mutation and natural selection.
The thing about a computer instantiation of evolution like AVIDA is that you can check back every lineage and examine the fitness of all precursors. Not only that, but you can choose your own environment, and how much selecting it does. There are some really key findings:
- Irreducibly complex functions evolve, where IC is defined as “if you take something away, it breaks”
- Functions evolve via Irreducibly Complex pathways, i.e. pathways in which there are several neutral, or even quite steeply deleterious steps.
- The more complex functions do not evolve without selection.
- All the functions evolve via at least some neutral steps
- There are many pathways to each function
- Not all the functions are achieved in the same way.
The AVIDA findings are very good evidence that there is no reason, in principle, why irreducibly complex things can’t evolve, that some degree of natural selection aka heritable variance in reproductive success in the current environment is necessary for for certain features to evolve, but that not all steps need be advantageous – because of the role of drift, even quite sharply deleterious steps can prove key to the emergence of a complex function.
Then there’s the Lenski lab e-coli experiments. But let’s start with AVIDA, because it really does rebut the idea that Irreducible Complexity, whether defined as the property of a function or of a path to a function, is an in-principle bar to evolution by natural selection.
As the Natural Selection and Adaptation thread was getting rather long, I thought this one might give us an extension, and the Lenski is an important and relevant paper.
The first question I’d like to ask is whether they call what goes on in their program natural selection. Or do they reserve the use of natural selection for what goes on in the natural world, as, say, contrasted with an artificial world.
Or do folks here (not named Gregory) just naturally reject the artificial/natural distinction?
If you are doing it in a program, then it is a simulation. So the better question would be to ask how realistic is the simulation.
I’m hoping for ‘my flight simulator doesn’t even leave the room’.
Neil,
No, AVIDA is not a simulation.
From the abstract:
ETA: And from the paper itself:
What goes on in their programme is natural selection. The selection parameters are set at the beginning, and form the “environment” in which the virtual organisms evolve – an environment that offers “resources” to those organisms that find themselves capable of exploiting them, in the form of “energy” with which to reproduce as well as “foraging” for “food”.
The “foraging” takes the form of performing logic functions. If they perform a logic function they get “food”. The fancier the logic function they perform, the more food they get. A biological equivalent might be that the faster a creature can move, the more likely it is to get to the food before the competition. And there are lots of features that could help the creature move fast – same here. Also lots of different ways of constructing those features, just as there are in biology (e.g. batwings and bird wings).
Well, I didn’t call it a “simulation” for good reason – I called it an “instantiation”. It isn’t “like” Darwinian evolution. It’s an instantiation of it – the virtual organisms really evolve, and they really end up with neat little programs that nobody wrote that perform actual logic functions.
The issue I guess is to what extent the instantiation resembles the putative instantiation in biology.
It seems to. It’s a non-sexually reproducing system, so more like bacteria than us. But no HGT.
So it lacks some advantages of real biological systems, in that the genes can’t propagate independently of the genome in which they occurred. HGT does this to some extent in bacteria, and sexual reproduction does it for the rest of us.
Elizabeth, do you think perhaps some of the problems people have with how you communicate your ideas has something to do with the over-use of scare quotes?
Why is it that natural selection requires no scare quotes but “environment” and “resources” and “energy” and “foraging” and “food” all require scare quotes?
Might one at least be allowed to be skeptical of your claim that what goes on the the program really is “natural selection” when all the words you have to use to defend that claim have to be qualified in a way that I assume you mean to convey they should not be taken literally?
But we should take your use of natural selection literally?
I agree with Neil and disagree with Elizabeth. Avida is a simulation.
True, Avida stands on its own for what it is and does:
And:
But that does not mean it isn’t a simulation.
If Avida is not a simulation then it’s hard to see how it is even relevant to the debate over ID. We all agree Avida was intelligently designed, right?
Behe’s claims were about natural biological systems.
So my next question is, how is Avida unlike natural biological systems?
Is there genotype/phenotype distinction and if so does it reflect the gene/protein distinction and the genetic code?
It would appear to consist of a set of fixed and finite “genes.” Fitness appears to be pre-determined and the environment appears to be fixed.
Evolution will occur regardless of whether there is mutation.
Evolution will occur regardless of whether there is differential fitness (competition).
I can’t imagine that that makes a difference, Mung, but I put them in quotes because they aren’t literally “food” etc, because they are numbers in a computer.
Food is what I have in my cereal bowl right now. “Food” in AVIDA is some value added to a virtual organism aka an “organism” when it executes a logic operation.
http://evoinfo.org/minivida/
http://bio-complexity.org/ojs/index.php/main/article/view/BIO-C.2010.3
http://evoinfo.org/papers/2009_EvolutionarySynthesis.pdf
Yes, hard to see how it’s related. Certainly nobody else involved in ID seems to think it’s relevant.
http://bio-complexity.org/ojs/index.php/main/article/viewFile/BIO-C.2014.1/BIO-C.2014.1
http://www.iscid.org/papers/Anderson_BitByte_020305.pdf
Seems bio-complexity and iscid disagree with Mung that Behe’s claims can be tied to Avida. But no real surprise there.
In my response below, anything in quotes is a metaphor or analog.
As said, there is no mechanism for a genetic sequence to pass from individual to individual independently of the rest of that individual’s genome. So they are more like bacteria than, say, rats – but there is no HGT either.
So inheritance is strictly linear, with no recombination of genomes.
Yes, there is a genotype/phenotype distinction, and there is a genetic code with 26 possible “letters” (analogous to codons), each being a single instruction (analogous to an amino acid). The genome is circular, and the instructions are carried out in sequence except when a sequence of instructions amounts to an instruction to “jump” to a more distant instruction. So that is quite close to the codon-protein transcription/translation system, including regulatory sequences that control what gets read next. Instead of proteins and enzymes being the output, however, the output is logic operations. Not all sequences of instructions will do anything at all. The one capacity built in at the start (because this does not emulate OoL) is the capacity to copy itself. The sequence of instructions to copy itself is part of the genome. When it does so, it does so without 100% fidelity, in other words some “offspring” (copies of a virtual organism) have genomes that are not identical to the genome of their parent.
Not sure what you mean, but if a “gene” is a sequence of instructions that performs an operation, then no, the genes are not “fixed”. They start with only one “gene” – the gene for copying themselves, sometimes with mutations. The mutations can involve point mutations (changing one instuction for another), insertions (adding a stretch of instructions) and deletions (removing a set of instructions). I don’t think there is an upper limit to genome length.
During a run, the virtual organisms compete for resources. As in life, therefore, the biological environment changes as the population changes – a genome that gives you a resource-accessing feature that your peers don’t have in one generation and is therefore “advantageous” may not give you the edge any more when your peers have both it and more. However, the non-biological environment doesn’t change – it is set at the start. Certain logic operations yield a certain amount of “food”, and in addition, organisms with longer genomes get more “food”. So the environment, as in life, selects for certain features. In this case it is size and capacity to perform logic operations.
The size and capacity to perform logic operations is the phenotype. The circular set of instructions is the genome.
It’s a simulation in the sense that it is like biology. It is an instantiation in the sense that real evolution goes on in it.
And it is highly relevant to Behe’s claims as it directly tests them.
The virtual organisms are not intelligently designed, apart from the initial population, which has a single function, namely the capacity to replicate with randomly introduced variance. The capacities they evolve are not intelligently designed – they are the result of the same process that evolutionary theory postulates is responsible for the evolution of complex adaptive features. So AVIDA is a directd test of whether Darwin’s proposed mechanism does in fact result in complex features.
The environment is intelligently designed because you can’t build an artifical environment in a computer without using some human intelligence. But the computer environment is the analog of the real environment. If ID proponents want to argue that the earth’s non-biotic environment must have been intelligently designed so as to favour the evolution of dinosaurs, humans, trees, etc, that’s fine. But it wasn’t Behe’s argument, although it’s getting close to Dembski’s current argument, and indeed the “fine-tuning” argument.
It’s Behe’s claim that is tested in AVIDA.
Behe’s claim was that Irreducibly Complex systems couldn’t evolve. His argument was not specific to biological systems, but obviously is only applicable to self-reproducing populations, and the only two examples we know of so far of those are biological systems and in silico systems.
Oh, another difference is that there are no epigenetic effects.
OMagain, you missed one:
http://www.tbiomed.com/content/8/1/9
Do they say where the line from “not complex” to “complex” is crossed?
Mung,
Only up to the point where variation has been eliminated, or (more rarely) a frequency-dependent stable point.
This is true. Dennett is wrong (or rather, those aren’t the only conditions in which evolution will occur. The wider quote may clarify that, for all anyone can tell from your unreferenced snippet).
There is no line. All are polygeneic so all are complex in the sense that more than one mutation is required to get to any of them.
They made it hard for themselves. None of the functions can be achieved by a single mutation. All are therefore IC by at least one of Behe’s definitions.
I reject it too. It is merely semantical. Human beings have been naturally selecting organisms for millenia before anyone ever heard of Darwin. That’s why the apples in your local grocery store are huge and full of sugar compared to their ancestors. That’s why the local dairy cow probably couldn’t even live in the wild because it’s basically become a big fat milk-producing organic machine.
These results were achieved centuries before Darwin, and certainly even longer before anyone ever head of genetics. Nobody forced the mutations to happen, something was just picked for breeding because humans “wanted more of it”. This doesn’t even have to be a choice done with any foresight or understanding. When you pick a “cute dog” to take home with you, you don’t have to plan for it to get puppies sometime in the future. Something appealed to you and the result down the line is that it gets to multiply above it’s “competition” without anyone ever having to sit and reason these events out beforehand.
The distinction between natural and artificial selection is entirely semantical.
No, but neither does any creationist. Ever.
In fact they can’t even give a way to reliably measure the level of complexity of anything, at all. It’s just a fancy term used because it, well because it sounds fancy. Omg, it’s complex, praise the LORD!
Not really no. Mostly because to simulate an organism to such a detail you capture the physics of molecular interactions at the gene-protein scale is simply beyond the capability current computer hardware.
Even if we could simulate a very reduced organism, with a few hundred genes, that would basically be it. It would still not be possible to simulate a population of them.
Also note that even at the highly reduced level of detail that simulations like avida run at, they still deal only with relatively small population sizes in their thousands to a few tens of thousands. In your gut right now there’s probably close to a trillion bacteria.
So it’s not that the “unlikeness” of the Avida simulation is out of some nefarious attempt to circumvent important biological details, it is simply out of necessity. We still don’t have the hardware that can capture and simulate something that looks so much like the real world that you can extract analogies to the physics that takes place when proteins are synthesized, fold and interact with each other, the genome and other parts of the intracellular environment.
So somewhere along the way, you just have to make compromises.
It would be trivial to alter the simulation so the fitness values are fluctuating, either around a mean, completely randomly or according to some function.
But the result of that would not be interesting because we can already predict what will happen: Over any arbitrary length of generations, the “population” will on average be in the process of becoming more fit. That’s just how selection works.
But nevertheless, such simulations have been run and we basically just see what I said. As the previous “environment” starts changing into a new one, the population will move to the new “peak” over a few generations while moving away from the old one. See this nice blogpost by Bjørn Østman about evolution in dynamic fitness landscapes:
http://pleiotropy.fieldofscience.com/2014/06/video-visualizing-coevolution-in.html
Rumraket,
It isn’t even out of necessity, because AVIDA was never intended to be a simulation of biological evolution.
The goal was to study evolution — actual evolution — in a non-biological context:
AVIDA is not a simulation. It shares some characteristics with biological evolution, but it is not a simulation of biological evolution. It is a real instance of evolution in a non-biological “world”.
Who says these math things are complex? they are not complex nut merely about memory operation.
biology is complex. its not simple math memory stuff.
I don’t see why this is a good analogy with biology.!!
How could this computer stuff fail? it must progress as if there is a programmed direction. From a creator.
Biology would not have a direction.
IC does mean it broke it don’t work and so reproduce.
Math stuff here is not under such stress.
The computer stiff here is not creating new complex things and when done its still not copmplex but a mere addition of general equations. No novel ideas as biology would need if evolved.
For goodness’ sake, please don’t let’s get bogged down over whether AVIDA is a simulation or an emulation or an instantiation,
It’s all of those, in some sense or other.
The point is that the principle on which the virtual organisms evolve is exactly the principle Darwin proposed for biological evolution. It’s a test of that principle.
Not only that, but it’s a test of Behe’s hypothesis, which is that if a thing is IC it can’t evolve.
Behe’s definition of IC does not require that the thing be biological. His famous example is of an IC thing is a mousetrap.
The virtual organisms in AVIDA are, in Behe’s terms, mousetraps.
And they evolve, in precisely the manner Behe says they can’t – by means of lineages that include variants that are no more fit, and in some cases, substantially less fit, than their parents.
That is the key finding, and that is the issue that I have yet to see any ID proponent, or Behe advocate, address.
EL said:
Perhaps you can provide the quotes from the paper that support this extrapolation? The only thing I can find that you might be talking about is this:
Is that what you are referring to?
Well, that’s one example. But AVIDA is open source and there are lot of examples of people trying different combos, e.g. not rewarding any at functions at all (zero selection). I can’t remember whether occasionally one of the simpler ones makes a brief appearance, but EQU never in any run I am aware of.
However, rewarding EQU is a more stringent test – because there is actually a reward for EQU. And while EQU evolves reliably if there are some intermediate rewarded steps, it never does if there aren’t.
So the demonstration does three important things:
It shows that natural selection makes the evolution of EQU possible.
It shows that even though EQU is IC, it still evolves.
It shows that there can be a substantial number (but not an unlimited) number of neutral or deleterious steps between EQU and it’s predecessors – indeed, in the case of EQU, those neutral and deleterious steps are necessary.
Therefore EQU is IC by both the definitions Behe has offered: if you remove any part it breaks (definition one: IC feature); the only paths to it involve many neutral, and even deleterious, steps (definition two – IC pathways). And yet it evolves.
It doesn’t evolve without NS.
It does evolve with.
It evolves with, despite being IC, and despite the only evolutionary pathways to it including substantial sequences of neutral or deleterious steps.
EL,
I think there may be a problem with what you are calling “without NS”. The paper says that when the simpler variations were “not rewarded”, they were “not preserved”. Why weren’t they preserved, if there is no NS?
Natural selection only “preserves” a variant in the sense that it doesn’t kill it off while it kills off its competitors. Without Natural Selection (no dying, no competition), all variations are preserved, because no variation dies and all variations continue to reproduce, which means they keep producing exact and inexact duplicates of themselves, and their descendents keep producing exact and inexact duplicates of themselves. Thus, all variations are preserved.
I don’t think that set of parameters was covered by this experiment. Perhaps I’m wrong and you can provide a quote that indicates otherwise?
EL said:
Again, this depends on what you mean by “evolve”. If “evolve” includes Natural Selection by definition, then okay, it doesn’t “evolve”. What it would do, however, is eventually produce every conceivable feature, including a massive amount of features likely unavailable if constrained by natural selection. And, RM by itself would ikely produce them at least as fast as when any version of Natural Selection is used to filter outcomes.
William,
You’re missing a key point, which is that resources are finite. What you say would be true with infinite resources, but it is decidedly false when resources are finite.
Think of Dawkins’ Weasel running with and without selection. You get convergence quickly when there is selection. Without selection, you will be waiting many lifetimes before you see “METHINKS IT IS LIKE A WEASEL”.
No, evolve is not restricted to evolution by natural selection. Things can also evolve by drift. In fact, drift is a major player in the AVIDA example.
And we can test your claim that “RM by itself would ikely produce them at least as fast as when any version of Natural Selection is used to filter outcomes”. It’s tested in that paper. It doesn’t.
Turn off NS, and EQU doesn’t evolve in the length of time of any run yet done. Turn it on and it does, reliably.
Would you like to bet on this? Perhaps a bottle of Whiskey?
No. That is a very special and non-existent case of “No NS”.
There are real life cases of “No NS” i.e. where genotype is totally uncorrelated with probability of reproduction.
And that can be instantiated in AVIDA – all organisms have the same probability of reproducting, regardless of their genotype. When you do that, EQU doesn’t evolve. Add NS – i.e. make those organisms that can perform a logic function extra “food” and thus increase their chances of reproducing, and EQU reliable evolves.
Which is exactly analogous to biology, where an extra inch of leg or spot of camouflage can improve your chances of eating or not being eaten and thus of leaving progeny.
He has already quoted the passage where he loses the bet.
For pete’s sake don’t let’s get sidetracked by WEASEL.
Keiths said:
Finite resources are part of natural selection. So, you agree with me.
EL: You didn’t answer my question. Without NS, why aren’t the simpler variations preserved in the experiment? Without NZ, there is no dying. There is no competition for resources. There is no diminished reproduction
EL said:
No. You haven’t explained why the earlier, simpler variations are not preserved, if there is no natural selection. Where did they go?
The timer says I have time to edit the comment (change NZ to NS), but when I click on edit, it says I don’t have permission to edit. What’s up?
And now, suddenly, all my timers dropped to zero. Except this one. Even though I just posted the others a few minutes ago. Strange.
William J. Murray,
And now, suddenly, all my timers dropped to zero. Except this one. Even though I just posted the others a few minutes ago. Strange.
Natural selection
No natural selection =/= No death.
It just means that death, or non-reproduction is unrelated to genotype.
WordPress has glitches. Everything does – given the multiplicities of browsers and operating systems it’s amazing the interactive internet works at all.
EL said:
I didn’t claim it was “equal to”. I said, it’s an aspect of NS. It is. As is resource limitation. If a thing dies, it no longer reproduces or generates variations of its particular feature code, which would reduce the fecundity of that particular line – especially if it dies soon after being generated. Whatever caused its death, for whatever reason, would be an aspect of natural selection.
No. If its death was random with respect to it genetic makeup; if it was just ‘bad luck’ it wouldn’t be natural selection. NS is defined as the change of the genetic makeup of a population due to the differential fitness of different genotypes. Evolution includes NS but also includes drift: Drift is when the genetic makeup of a population changes over time due to the random nature of survival and reproduction.
Thank you RodW.
There can be dying without NS. So what “happened” in the NS scenario is that they died. Shit happens sometimes.
No prob Elizabeth! I’m shocked that my comment was actually visible!
Any chance we could shift the discussion to the only ‘official’ critique made of AVIDA? .. the paper by Marks and Dembski which essentially said that the appearance of IC, info and whatever else is really an illusion because information was smuggled in with the program itself.
Ah, I knew the computer models were lacking something! Actual meteor storms.